Pith. sign in

REVIEW 2 major objections 6 minor 31 references

HourGlass reconstructs the hours between 6-hourly data-driven weather forecasts as skillful, temporally coherent probabilistic trajectories with realistic small-scale structure.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-14 05:27 UTC pith:2BFBSEEA

load-bearing objection Solid operational methods paper: probabilistic CRPS temporal downscaling trained on continuous NWP trajectories, verified on independent SYNOP, with code released. the 2 major comments →

arxiv 2607.11457 v1 pith:2BFBSEEA submitted 2026-07-13 physics.ao-ph

HourGlass: A probabilistic data-driven temporal downscaler for global and regional weather forecasting

Magnus Sikora Ingstad , Mariana C. A. Clare , Olav Ersland , Vera Gahlen , H{\aa}vard Homleid Haugen , Oph\'elia Miralles , Even M. Nordhagen , Thomas N. Nipen
show 11 more authors
Ivar A. Seierstad John Bj{\o}rnar Bremnes Michael Maier-Gerber Zied Ben Bouall\`egue Harrison Cook Christian Lessig Gert Mertes Cathal O'Brien Florian Pinault Ana Prieto Nemesio Matthew Chantry
This is my paper
classification physics.ao-ph
keywords temporal downscalingdata-driven weather forecastsprobabilistic forecastingCRPShourly weather predictionensemble forecastinggraph neural networks
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Most leading data-driven weather systems still step only every six hours, yet operations and extreme-event work need hourly detail. Direct hourly forecasting piles up error and can learn unphysical time correlations, so HourGlass instead reconstructs the intermediate hours from the two bounding forecast states. It is trained probabilistically with continuous-ranked-probability losses plus extra penalties on window min, max, mean and successive differences, so uncertainty appears as spread rather than as spatial smoothing, and successive hours evolve coherently. Training targets are continuous segments of numerical weather prediction forecast trajectories rather than reanalysis, avoiding assimilation-window jumps that earlier methods absorbed. Applied both globally and regionally, the models keep the skill of their parent 6-hourly systems, produce spectra of temporal increments that match numerical benchmarks, and track storms and organised convection in a physically consistent way, while intense hourly precipitation extremes remain underestimated.

Core claim

A probabilistic temporal downscaler trained with almost-fair CRPS plus window-wise min/max/mean and difference terms on continuous NWP forecast segments can fill the hours between 6-hourly data-driven forecasts while preserving upstream skill, realistic small-scale spatial variability, and coherent physical evolution, rather than producing the temporally inconsistent smoothing of deterministic MSE-trained interpolators.

What carries the argument

Almost-fair CRPS on each output hour, augmented by the same score applied to min, max, mean and successive differences over the six-hour window (plus a regional spectral term), with shared latent noise so one forward pass yields a single coherent trajectory; residual connections force the final hour of prognostic fields to match the parent forecast.

Load-bearing premise

Continuous stretches of numerical weather prediction forecast trajectories teach the true physical hour-to-hour evolution without imprinting the parent model’s biases and spin-up so strongly that skill collapses when the downscaler is later applied to analysis-trained AI forecasts.

What would settle it

Hourly verification against surface stations would show larger errors or broken temporal-increment spectra inside each 6-hour window relative to cubic-spline interpolation of the same parent forecast, or storm case studies would exhibit unphysical jumps or loss of frontal structure at mid-window.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Existing 6-hourly data-driven forecast systems can supply operational hourly products without being retrained for hourly steps.
  • Skill gains that come from analysis-trained parent models (for example better 2 m temperature) can be carried into the intermediate hours even though the downscaler itself never saw analyses as targets.
  • Deterministic or MSE-trained temporal interpolators become inferior defaults whenever small-scale spatial realism and coherent storm timing matter.
  • Each ensemble member can be downscaled independently, preserving a probabilistic hourly product at modest extra cost.
  • Improvements to the 6-hourly parent forecasts, especially of intensity extremes, are expected to translate directly into better hourly fields.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same boundary-anchored residual design should generalise to other fixed temporal gaps (for example 3-hourly to hourly) provided the parent anchors remain skilful.
  • Diagnostic precipitation’s over-confidence near the window edges suggests multi-scale or boundary-aware probabilistic losses will be needed before hourly rain products match operational needs.
  • Training only on post-spin-up forecast segments may systematically under-represent the first hours after analysis, limiting direct use for nowcasting-style applications.
  • Once upstream AI models reduce dry bias and raise extreme rain rates, HourGlass should inherit those gains at hourly resolution with little additional training cost.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

2 major / 6 minor

Summary. The manuscript introduces HourGlass, a probabilistic GNN-based temporal downscaler that reconstructs intermediate hourly states between two 6-hourly forecast anchors. Models are trained with almost-fair CRPS plus min/max/mean/difference aggregate terms (and a regional spectral term for Bris) on continuous NWP forecast segments (IFS and MEPS) rather than reanalysis, to avoid assimilation-window jumps. Two applications are presented: global AIFS-HourGlass (applied to AIFS-Single and AIFS-ENS) and regional stretched-grid Bris-HourGlass. Verification against independent 2025 SYNOP stations, spectral diagnostics of fields and temporal increments, loss ablations, cubic-spline and NWP baselines, and case studies of Storm Amy and US Southern Plains convection support the claim that skill of the upstream 6-hourly systems is retained while producing temporally coherent hourly trajectories with improved small-scale spatial variability. Precipitation extremes remain underestimated, which the authors acknowledge as a shared data-driven limitation.

Significance. If the results hold, HourGlass supplies a practical, operationally relevant bridge from the dominant 6-hourly data-driven forecast paradigm to the hourly products required by many applications, without the error accumulation of direct hourly forecasting or the smoothing of deterministic temporal interpolators. Strengths include independent SYNOP verification for 2025, spectral analysis of increments (Fig. 10), explicit loss ablations (Figs. 6–8), residual connections that leave 6-hour anchors unchanged, training on forecast trajectories to avoid ERA5 jumpiness (Figs. 4–5), dual global/regional demonstration, and open code in the Anemoi framework. These elements make the contribution concrete and reusable rather than purely methodological.

major comments (2)
  1. §3.2 and Fig. 14: When Bris-HourGlass is driven by Bris (or by MEPS analysis), mean hourly precipitation exhibits clear mid-window relaxation toward the training distribution and a dry bias inherited from the upstream model near the 6-hour anchors. The paper correctly notes that 6-hourly accumulated precipitation CRPS improves, but the residual boundary dependence and spin-up imprint remain load-bearing for the claim of seamless hourly products. A short quantitative summary of how much of the mid-window skill gain is pure spread versus genuine error reduction (already hinted at by Fig. 13) would strengthen the operational interpretation.
  2. §2.2–2.3 and Fig. 5: The decision to train exclusively on IFS/MEPS forecast segments (lead times 12–29 h / 6–23 h) is well motivated by ERA5 jumpiness, yet the manuscript does not fully quantify how much parent-NWP bias or spin-up structure is transferred into the downscaler when it is later applied to analysis-trained AI forecasts. Fig. 5 shows comparable RMSE for ERA5-only vs IFS-only training on a few variables, but a parallel diagnostic for precipitation intensity distributions (analogous to Fig. 8) and for the regional model would make the weakest assumption more transparent.
minor comments (6)
  1. §2.1: The switch from a pure transformer processor (prior AIFS) to a graph transformer is stated but not motivated; a sentence on why this choice was made for multi-time-step decoding would help.
  2. Table 1: Asterisks and daggers for diagnostic vs prognostic variables and for global-only fields are dense; a short legend or footnote clarifying which variables are residual-connected would improve readability.
  3. Fig. 2 caption: “Hours 4-6 share the same random noise” is useful; stating explicitly that the same noise seed is used across the ablation would make the visual comparison clearer.
  4. §3.1: The small RMSE jumps attributed to diurnal observation density are plausible; a brief note on the number of stations per hour (or a supplementary plot) would remove residual ambiguity.
  5. Code availability: The GitHub link points to a feature branch; a commit hash or release tag would improve long-term reproducibility.
  6. Typographical: “availabilty” in the Code availability heading; “Code availabilty” should be corrected.

Circularity Check

0 steps flagged

No significant circularity; skill-preservation and coherence claims rest on external SYNOP/MRMS verification, with residual anchors left unchanged by design and openly discounted.

full rationale

HourGlass is trained on continuous NWP forecast segments (IFS lead times 12–29 h; MEPS 6–23 h) with almost-fair CRPS plus min/max/mean/difference aggregate terms, then applied to independent AIFS/Bris 6-hourly states and scored against SYNOP stations (2025) and MRMS QPE. The residual skip connection (Sec. 2.1) forces prognostic fields at the t6 anchor to equal the upstream forecast by construction, so 6-hourly skill is preserved tautologically; the authors explicitly state that “the scores at every 6th hour are not pertinent” and evaluate only the intermediate hours, increment spectra (Fig. 10), 24 h extrema (Fig. 15), and case-study evolution. Self-citations supply the upstream models (AIFS, Bris) and the CRPS recipe; they are not uniqueness theorems that force the downscaler result. No parameter fitted to one quantity is re-labelled a prediction of a closely related quantity, and precipitation biases inherited from the training distribution are documented rather than hidden (Fig. 14, Sec. 3.2). The central claims therefore stand on external benchmarks and are not circular.

Axiom & Free-Parameter Ledger

4 free parameters · 3 axioms · 1 invented entities

The paper is an empirical ML-methods contribution. Its load-bearing choices are architectural and loss-design decisions plus the decision to treat NWP forecast trajectories as ground truth for temporal evolution. No new physical entities are postulated; free parameters are the usual ML hyper-parameters and loss weights.

free parameters (4)
  • almost-fair CRPS epsilon = 0.025
    Set to 0.025 following Lang et al. (AIFS-CRPS); controls the fairness correction in the ensemble CRPS loss.
  • spectral loss weight lambda_f (Bris) = 0.15
    Relative weight of the regional FFT CRPS term; chosen as 0.15.
  • learning rates and iteration counts = various (see §2.4)
    Cosine schedules with effective LRs 8e-4 (AIFS), 3.2e-3 then 6.4e-4 (Bris stages), 200k/50k iterations; standard but free hyper-parameters that affect final skill.
  • variable tendency weights w_i = 1/sigma(Delta x_i) = data-derived
    Per-variable loss weights derived from training-set tendency statistics; re-scales the multi-variable CRPS.
axioms (3)
  • domain assumption Continuous segments of operational NWP forecasts (IFS, MEPS) provide temporally consistent and sufficiently realistic hourly targets for learning physical evolution.
    Stated in §2.3; replaces ERA5 because of 4D-Var jumps. If the NWP trajectories contain systematic spin-up or bias, the downscaler inherits them (illustrated for precipitation in Fig. 14).
  • ad hoc to paper Almost-fair CRPS plus min/max/mean/difference aggregates is an adequate surrogate for temporal coherence and small-scale spatial variability.
    Introduced in §2.2; ablation in Figs. 2, 6–8 supports the choice but the functional form is a design decision, not derived from first principles.
  • domain assumption A residual connection that forces the final (t+6) prognostic state to equal the input forecaster state is valid and does not harm intermediate-hour skill.
    §2.1; ensures the downscaler cannot degrade the parent forecast at anchor times.
invented entities (1)
  • HourGlass temporal downscaler (AIFS-HourGlass / Bris-HourGlass) no independent evidence
    purpose: Probabilistic multi-hour reconstruction of intermediate states between two forecast anchors.
    The named method and its two instantiations; evaluated empirically, no independent physical existence claimed beyond the trained networks.

pith-pipeline@v1.1.0-grok45 · 23011 in / 3138 out tokens · 31945 ms · 2026-07-14T05:27:03.050004+00:00 · methodology

0 comments
read the original abstract

Many forecast applications require high frequency temporal resolution, yet most state-of-the-art data-driven weather forecasting systems operate at 6-hourly resolution. Although direct hourly forecasting is possible, it suffers from error accumulation and temporal inconsistency. We introduce HourGlass, a probabilistic data-driven temporal downscaling method that reconstructs the evolution between forecast states. HourGlass is trained using variants of the continuous ranked probability score (CRPS) preserving small-scale spatial variability while encouraging temporal consistency. Unlike existing deterministic temporal downscaling approaches, which tend to produce overly smooth fields, HourGlass generates realistic probabilistic forecasts. Training on forecast trajectories rather than reanalysis or analysis data also avoids the temporal inconsistencies present in datasets used by previous methods. We evaluate HourGlass in two settings: AIFS-HourGlass, applied globally to ECMWF's AIFS-Single and AIFS-ENS forecast systems, and Bris-HourGlass, applied regionally to MET Norway's high-resolution stretched-grid ensemble model, Bris. Verification against observations shows that both models retain the skill of their underlying forecasting systems while producing temporally coherent hourly forecasts with realistic small-scale variability. Case studies demonstrate physically consistent evolution during rapidly developing weather events, including extratropical cyclones and organised convection. Hourly precipitation remains challenging: HourGlass improves the spatial realism of precipitation fields but still underestimates the most intense extremes, a common limitation of data-driven weather forecasting models. These results demonstrate that HourGlass effectively bridges the gap between 6-hourly data-driven forecasts and the hourly products required for operational regional and global forecasting.

Figures

Figures reproduced from arXiv: 2607.11457 by Ana Prieto Nemesio, Cathal O'Brien, Christian Lessig, Even M. Nordhagen, Florian Pinault, Gert Mertes, H{\aa}vard Homleid Haugen, Harrison Cook, Ivar A. Seierstad, John Bj{\o}rnar Bremnes, Magnus Sikora Ingstad, Mariana C. A. Clare, Matthew Chantry, Michael Maier-Gerber, Olav Ersland, Oph\'elia Miralles, Thomas N. Nipen, Vera Gahlen, Zied Ben Bouall\`egue.

Figure 1
Figure 1. Figure 1: Example of how a forecasting and temporal downscaling model can be used in inference. Pictured at t and [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Hourly precipitation forecast off the west coast [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Construction of forecast datasets for training. A training sample must stay within the boundaries marked in [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Schematic illustration of jumpiness introduced by 4D-Var assimilation cycling. Stitching forecasts from [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: RMSE of hourly forecasts produced by AIFS [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: RMSE of hourly forecasts produced by AIFS [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 8
Figure 8. Figure 8: Contribution of each rain rate to the total accumulated precipitation (precipitation * f(precipitation) where f(.) [PITH_FULL_IMAGE:figures/full_fig_p009_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Comparison of spectra of precipitation at a lead [PITH_FULL_IMAGE:figures/full_fig_p009_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Comparison of spectra of temporal increment of 10m u-wind velocity of AIFS-HourGlass Single and IFS. [PITH_FULL_IMAGE:figures/full_fig_p010_10.png] view at source ↗
Figure 12
Figure 12. Figure 12: Fair CRPS for hourly and 6-hourly accumulated [PITH_FULL_IMAGE:figures/full_fig_p011_12.png] view at source ↗
Figure 11
Figure 11. Figure 11: Fair CRPS of hourly forecasts produced by [PITH_FULL_IMAGE:figures/full_fig_p011_11.png] view at source ↗
Figure 13
Figure 13. Figure 13: Single-member RMSE and standard devia￾tion (unbiased root mean variance) of 1-hour accumulated precipitation for the MEPS ensemble forecast and Bris￾HourGlass. Verification is performed against SYNOP for 2025. 11 [PITH_FULL_IMAGE:figures/full_fig_p011_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: 1-hour accumulated precipitation averaged over [PITH_FULL_IMAGE:figures/full_fig_p012_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Fair CRPS for 24-hour maximum of 2m tem [PITH_FULL_IMAGE:figures/full_fig_p012_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Meteogram for Kjevik airport (SN39040), Nor [PITH_FULL_IMAGE:figures/full_fig_p012_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Spatio-temporal evolution of (a) hourly precipitation rate and (b) 10 m wind speed and mean sea level [PITH_FULL_IMAGE:figures/full_fig_p014_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: Spatio-temporal evolution of (a) hourly precipitation rate and (b) 10m wind speed and mean sea level [PITH_FULL_IMAGE:figures/full_fig_p015_18.png] view at source ↗
Figure 19
Figure 19. Figure 19: Time series of precipitation and atmospheric [PITH_FULL_IMAGE:figures/full_fig_p016_19.png] view at source ↗
Figure 20
Figure 20. Figure 20: Spatial distribution of hourly total precipitation rate (in mm/h) from observations (MRMS QPE), and [PITH_FULL_IMAGE:figures/full_fig_p017_20.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

31 extracted references · 4 canonical work pages

  1. [1]

    Machine learning for precipitation nowcasting from radar im- ages, 2019

    Shreya Agrawal, Luke Barrington, Carla Bromberg, John Burge, Cenk Gazen, and Jason Hickey. Machine learning for precipitation nowcasting from radar im- ages, 2019. URL https://arxiv.org/abs/1912 .12132

    2019

  2. [2]

    The rise of data-driven weather forecasting: A first statisti- cal assessment of machine learning–based weather forecasts in an operational-like context

    Zied Ben Bouallègue, Mariana CA Clare, Linus Mag- nusson, Estibaliz Gascon, Michael Maier-Gerber, Martin Janoušek, Mark Rodwell, Florian Pinault, Jes- per S Dramsch, Simon TK Lang, et al. The rise of data-driven weather forecasting: A first statisti- cal assessment of machine learning–based weather forecasts in an operational-like context. Bulletin of the...

    2024

  3. [3]

    Pangu-weather: A 3d high- resolution model for fast and accurate global weather forecast, 2022

    Kaifeng Bi, Lingxi Xie, Hengheng Zhang, Xin Chen, Xiaotao Gu, and Qi Tian. Pangu-weather: A 3d high- resolution model for fast and accurate global weather forecast, 2022. URL http://arxiv.org/abs/22 11.02556

    2022

  4. [4]

    FuXi: a cascade machine learning forecasting system for 15-day global weather forecast

    Lei Chen, Xiaohui Zhong, Feng Zhang, Yuan Cheng, Yinghui Xu, Yuan Qi, and Hao Li. FuXi: a cascade machine learning forecasting system for 15-day global weather forecast. npj Climate and Atmospheric Science, 6(1):190, 2023. ISSN 2397-

    2023

  5. [5]

    URL http s://www.nature.com/articles/s41612-023 -00512-1

    doi:10.1038/s41612-023-00512-1. URL http s://www.nature.com/articles/s41612-023 -00512-1

    doi:10.1038/s41612-023-00512-1

  6. [6]

    URL https://www.ecmwf.int/ en/about/media-centre/news/2024/anemoi -new-framework-weather-forecasting-bas ed-machine-learning

    ECMWF, 2024. URL https://www.ecmwf.int/ en/about/media-centre/news/2024/anemoi -new-framework-weather-forecasting-bas ed-machine-learning

    2024

  7. [7]

    IFS Documentation CY49R1 - Part III: Dynamics and Numerical Procedures, chapter 3

    ECMWF. IFS Documentation CY49R1 - Part III: Dynamics and Numerical Procedures, chapter 3. ECMWF, 2024. doi:10.21957/D04FB7A27E. URL https://www.ecmwf.int/bibcite/referenc e/81625

    doi:10.21957/d04fb7a27e 2024

  8. [8]

    IFS Documentation CY49R1 - Part IV: Physical Processes, chapter 4

    ECMWF. IFS Documentation CY49R1 - Part IV: Physical Processes, chapter 4. ECMWF, 2024. doi:10.21957/C731EE1102. URL https://www. ecmwf.int/bibcite/reference/81626

    doi:10.21957/c731ee1102 2024

  9. [9]

    The operational forecast process at metcoop

    Reima Eresmaa, Ulf Andrae, Lars Berggren, John Bjørnar Bremnes, Carl Fortelius, Inger-Lise Frogner, Erik Gregow, Rafael Grote, Martin Ridal, Ole Vignes, Ivar Ansper, Roohollah Azad, Jevg¯enijs Begens, Lisbeth Bergholt, Jostein Blyverket, Laila Daniel, Bjørg Jenny Engdahl, Helene Erlandsen, Mehdi Eshagh, Mikael Hasu, Mariken Homleid, Petter Ideström, Karl-...

    work page doi:10.1175/bams-d-25-0225.1 2026

  10. [10]

    HarmonEPS—The HARMONIE En- semble Prediction System

    Inger-Lise Frogner, Ulf Andrae, Jelena Bojarova, Al- fons Callado, Pau Escribà, Henrik Feddersen, Alan Hally, Janne Kauhanen, Roger Randriamampianina, Andrew Singleton, Geert Smet, Sibbo van der Veen, and Ole Vignes. HarmonEPS—The HARMONIE En- semble Prediction System. Weather and Forecasting, 34(6):1909–1937, December 2019. ISSN 1520-0434, 0882-8156. doi...

    doi:10.1175/waf-d-19-0030.1 1909

  11. [11]

    Singleton, Morten Ø

    Inger-Lise Frogner, Andrew T. Singleton, Morten Ø. Køltzow, and Ulf Andrae. Convection-permitting ensembles: Challenges related to their design and use. Quarterly Journal of the Royal Meteorological Society, 145(S1):90–106, September 2019. ISSN 0035-9009, 1477-870X. doi:10.1002/qj.3525. URL https://rmets.onlinelibrary.wiley.com/ doi/10.1002/qj.3525

    work page doi:10.1002/qj.3525 2019

  12. [12]

    Joseph G. Galway. The lifted index as a predic- tor of latent instability. Bulletin of the American Meteorological Society, 37(10):528–529, 1956. doi:10.1175/1520-0477-37.10.528

    work page doi:10.1175/1520-0477-37.10.528 1956

  13. [13]

    The ERA5 global reanaly- sis

    Hans Hersbach, Bill Bell, Paul Berrisford, Shoji Hirahara, András Horányi, Joaquín Muñoz-Sabater, Julien Nicolas, Carole Peubey, Raluca Radu, Di- nand Schepers, et al. The ERA5 global reanaly- sis. Quarterly Journal of the Royal Meteorological Society, 146(730):1999–2049, 2020

    1999

  14. [14]

    Forecasting Global Weather with Graph Neural Networks, February 2022

    Ryan Keisler. Forecasting Global Weather with Graph Neural Networks, February 2022. URL h t tp : / / a r x i v . o r g / a b s / 2 2 0 2 . 0 7 5 75. arXiv:2202.07575 [physics]

    Pith/arXiv arXiv 2022

  15. [15]

    Graphcast: Learning skillful medium- range global weather forecasting, 2023

    Remi Lam, Alvaro Sanchez-Gonzalez, Matthew Will- son, Peter Wirnsberger, Meire Fortunato, Ferran Alet, Suman Ravuri, Timo Ewalds, Zach Eaton- Rosen, Weihua Hu, Alexander Merose, Stephan Hoyer, George Holland, Oriol Vinyals, Jacklynn Stott, Alexander Pritzel, Shakir Mohamed, and Pe- ter Battaglia. Graphcast: Learning skillful medium- range global weather f...

    Pith/arXiv arXiv 2023

  16. [16]

    Simon Lang, Mihai Alexe, Matthew Chantry, Jesper Dramsch, Florian Pinault, Baudouin Raoult, Mariana C. A. Clare, Christian Lessig, Michael Maier-Gerber, Linus Magnusson, Zied Ben Bouallègue, Ana Pri- eto Nemesio, Peter D. Dueben, Andrew Brown, Florian Pappenberger, and Florence Rabier. AIFS - ECMWF’s data-driven forecasting system, 2024. URLhttp://arxiv.o...

    Pith/arXiv arXiv 2024

  17. [17]

    A multi-scale loss formulation for learning a proba- bilistic model with proper score optimisation, 2025

    Simon Lang, Martin Leutbecher, and Pedro Maciel. A multi-scale loss formulation for learning a proba- bilistic model with proper score optimisation, 2025. URLhttp://arxiv.org/abs/2506.10868

    Pith/arXiv arXiv 2025

  18. [18]

    Aifs-crps: en- semble forecasting using a model trained with a loss function based on the continuous ranked probability score

    Simon Lang, Mihai Alexe, Mariana CA Clare, Christopher Roberts, Rilwan Adewoyin, Zied Ben Bouallègue, Matthew Chantry, Jesper Dramsch, Peter D Dueben, Sara Hahner, et al. Aifs-crps: en- semble forecasting using a model trained with a loss function based on the continuous ranked probability score. npj Artificial Intelligence, 2(1):18, 2026

    2026

  19. [19]

    Modulated adaptive fourier neural oper- ators for temporal interpolation of weather forecasts,

    Jussi Leinonen, Boris Bonev, Thorsten Kurth, and Yair Cohen. Modulated adaptive fourier neural oper- ators for temporal interpolation of weather forecasts,

  20. [20]

    URL https://arxiv.org/abs/2410.189

  21. [21]

    arXiv:2410.18904 [physics.ao-ph]

    Pith/arXiv arXiv

  22. [22]

    An update to ECMWF’s machine-learned weather forecast model AIFS, 2025

    Gabriel Moldovan, Ewan Pinnington, Ana Prieto Nemesio, Simon Lang, Zied Ben Bouallègue, Jes- per Dramsch, Mihai Alexe, Mario Santa Cruz, Sara Hahner, Harrison Cook, Helen Theissen, Mariana Clare, Cathal O’Brien, Jan Polster, Linus Magnus- son, Gert Mertes, Florian Pinault, Baudouin Raoult, Patricia de Rosnay, Richard Forbes, and Matthew Chantry. An update...

    arXiv 2025

  23. [23]

    Re- gional data-driven weather modeling with a global stretched-grid

    Thomas Nils Nipen, Håvard Homleid Haugen, Magnus Sikora Ingstad, Even Marius Nordha- gen, Aram Farhad Shafiq Salihi, Paulina Tedesco, Ivar Ambjørn Seierstad, Jørn Kristiansen, Simon Lang, Mihai Alexe, Jesper Dramsch, Baudouin Raoult, Gert Mertes, and Matthew Chantry. Re- gional data-driven weather modeling with a global stretched-grid. Artificial Intellig...

    2025

  24. [24]

    URL https: //journals.ametsoc.org/view/journals/a ies/aop/AIES-D-25-0001.1/AIES-D-25-000 1.1.xml

    doi:10.1175/AIES-D-25-0001.1. URL https: //journals.ametsoc.org/view/journals/a ies/aop/AIES-D-25-0001.1/AIES-D-25-000 1.1.xml. Publisher: American Meteorological So- ciety Section: Artificial Intelligence for the Earth Systems

    doi:10.1175/aies-d-25-0001.1

  25. [25]

    High-resolution probabilistic data- driven weather modeling with a stretched-grid

    Even Marius Nordhagen, Håvard Homleid Haugen, Aram Farhad Shafiq Salihi, Magnus Sikora Ingstad, Thomas Nils Nipen, Ivar Ambjørn Seierstad, Inger- Lise Frogner, Mariana Clare, Simon Lang, Matthew Chantry, et al. High-resolution probabilistic data- driven weather modeling with a stretched-grid. arXiv preprint arXiv:2511.23043, 2025

    arXiv 2025

  26. [26]

    Graph-based neural weather prediction for lim- ited area modeling, 2023

    Joel Oskarsson, Tomas Landelius, and Fredrik Lind- sten. Graph-based neural weather prediction for lim- ited area modeling, 2023. URL http://arxiv.or g/abs/2309.17370

    Pith/arXiv arXiv 2023

  27. [27]

    Jaideep Pathak, Shashank Subramanian, Peter Har- rington, Sanjeev Raja, Ashesh Chattopadhyay, Morteza Mardani, Thorsten Kurth, David Hall, Zongyi Li, Kamyar Azizzadenesheli, Pedram Hassan- zadeh, Karthik Kashinath, and Animashree Anand- kumar. FourCastNet: A global data-driven high- resolution weather model using adaptive fourier neu- 19 HourGlass: A prob...

    Pith/arXiv arXiv 2022

  28. [28]

    Andersson, Andrew El-Kadi, Dominic Mas- ters, Timo Ewalds, Jacklynn Stott, Shakir Mohamed, Peter Battaglia, Remi Lam, and Matthew Willson

    Ilan Price, Alvaro Sanchez-Gonzalez, Ferran Alet, Tom R. Andersson, Andrew El-Kadi, Dominic Mas- ters, Timo Ewalds, Jacklynn Stott, Shakir Mohamed, Peter Battaglia, Remi Lam, and Matthew Willson. GenCast: Diffusion-based ensemble forecasting for medium-range weather, 2024. URL http://arxiv. org/abs/2312.15796

    Pith/arXiv arXiv 2024

  29. [29]

    The ALADIN System and its canonical model configurations AROME CY41T1 and ALARO CY40T1.Geoscientific Model Development, 11(1):257–281, January 2018

    Piet Termonia, Claude Fischer, Eric Bazile, François Bouyssel, Radmila Brožková, Pierre Bénard, Bogdan Bochenek, Daan Degrauwe, Mariá Derková, Ryad El Khatib, Rafiq Hamdi, Ján Mašek, Patricia Pottier, Neva Pristov, Yann Seity, Petra Smolíková, Oldˇrich Španiel, Martina Tudor, Yong Wang, Christoph Wittmann, and Alain Joly. The ALADIN System and its canonic...

    work page doi:10.5194/gmd-11-257-2018 2018

  30. [30]

    Multi-radar multi-sensor (mrms) quanti- tative precipitation estimation: Initial operating ca- pabilities

    Jian Zhang, Kenneth Howard, Carrie Langston, Brian Kaney, Youcun Qi, Lin Tang, Heather Grams, Yadong Wang, Stephen Cocks, Steven Martinaitis, Ami Arthur, Karen Cooper, Jeff Brogden, and David Kitzmiller. Multi-radar multi-sensor (mrms) quanti- tative precipitation estimation: Initial operating ca- pabilities. Bulletin of the American Meteorological Societ...

    doi:10.1175/bams- 2016

  31. [31]

    FuXi-2.0: Advancing machine learning weather forecasting model for practical ap- plications, 2024

    Xiaohui Zhong, Lei Chen, Xu Fan, Wenxu Qian, Jun Liu, and Hao Li. FuXi-2.0: Advancing machine learning weather forecasting model for practical ap- plications, 2024. URL http://arxiv.org/abs/ 2409.07188. 20

    Pith/arXiv arXiv 2024